Artificial Intelligence IRC Bots?

JonahC writes "dotcomma has an interesting programming challenge for people to create IRC bots with artificial intelligence and carry on conversations. With some development, good programmers and some luck, the bots should eventually interact with each other as if they are humans. "

I can see it now.

Throw them in #teens.

Bot #1: A/S/L

Bot #2: Where are you from?!@#!@?#

Bot #1: A/S/L

Bot #2: Where are you from?!@?#

etc

Been there - done that

About 5 years ago I wrote a very simple bot which matched regular-expressions and gave appropriate but generic responses. I realized after some time on IRC that conversations, particular those between males and "females", follow a pretty standard pattern. After 2 hours of hacking, I let my bot loose on the net. While some people spotted instantly that something was wrong, others spent literally hours talking to this simple program.
Inspired by this, I did consider creating a trigger based system (a trigger being a regexp to match, and an output sentence) which used a genetic algorithm to assess the success of each trigger and to make the best ones "breed". There was also a simple "flag" mechanism to act as a memory. While I acknowledge that such a program could never pass the Turing test, the important thing to remember is that this IRC challenge is not a Turing test as (I suspect) most of the "judges" will be unaware that they are talking to a piece of software and are therefore more likely to give the system the benefit of the doubt when it says something really stupid.
If any slashdotters are planning on entering this, email me, I will be happy to pass on my thoughts and knowledge.

--

Unlikely to be very successful

Being able to talk with each other 'as if they were human' would be equivalent to passing the Turing Test. And nobody has managed that, despite fifty years of trying.

Artificial Stupidity.

Considering the kind of human you normally encounter on IRC wouldn't it be more convincing to create artificial stupidity ?

Let the bot enter #linux and yell "HELLO ROOM" then proceed to ask complex and interesting questions in #windows95 ( after being kicked from #linux ). "The window on my car is stuck and some dolt at the shop says it's because I spilled glue in there. Could he be right ?"

Who knows this bot could even get opp status and monitor a few channels. oppbot* on Windows "HaXoR25 was kicked for straying off topic. 'Uptime' is not an allowed word here".

Do something fun for a change :)

I can see it now...

* HackerD00D has entered #warez4free
-HackerD00D- any1 have all 5 gigs of quake 4? will trade for entire backstreet boys mp3 collection
-ElizaBot- What do you feel about backstreet boys mp3 collection?
-HackerD00D- oh man, their the gr8est! there 40th reunion tour was the bomb.
-ElizaBot- So, you have feelings towards the bomb?
-HackerD00D- what, me? no way d00d! Get off my case!
* EschlonBot has entered #warez4free
* EschlonBot rudely grabs HackerD00D and locks him away.
* HackerD00D has left IRC (BUSTED!)
-EschlonBot- Thank you for your cooperation
* EschelonBot has left IRC (Obey the prime directives)
-ElizaBot- Tell me about your feels on obey the prime directives.

gNiall

There's an interesting little program with which you can converse. It's called gNiall, and you can find it at Freshmeat. You talk to it, and it looks for which words come next to which word and how often. So then it constructs sentences based upon these words. They're usually pretty nonsensical, but sometimes it'll reply something profound or really "on-topic" and it'll make you wonder. It's good fun to play with, although you can't feed it books yet to learn. Check it out!

Cheers!

Costyn.

As nifty as it sounds...

I am strongly against this idea for one simple reason.

Spammers will probably use this to better evade IRCops like myself who spend hours dealing with their stupid BS in the first place. pr0n, Long Distance, etc...

We already have a hard enough time weeding these idiots who flood channels, mass msg, and invite whole segments of lusers. Then of course, you have the zoom bots, the ones that join, get the channel list, ignore the IRCops, spam the rest and leave.

Add some AI to it, now, converrsation can ensue, possibly making it much easier to evade the policy of the network. Why stop at conversation? Lets teach the bots how to BNC when it gets an autokill or gline. For client based bots, how about we teach them to dial back in, to gain another IP. To change its ident, so it can evade any KLine/autokill/gline and keep right on spamming.

*sigh*

Want to invent something worthwhile? Streamline ircd for the next 10 years. Its starting to show its age. =)

The name is E.L.M.O

I have been experimenting with AI bots for a few years now and decided to write an AI irc bot. The project name is ELMO (it is a real acronym, good luck guessing it tho ;)) and it is pretty much finished since I reached the limits of mIRC. More programming would force me to write it in C. It is able to fool people for a long time, including experienced irc users (tested this with many people). It is not just an on 1:text... bot, this one learns, is able to build new sentences which haven't been preprogrammed, it is able to learn and pickup people's habits (which can cause funny situations sometimes), it drinks, it sleeps, it watches tv (you will laughing when you find out what his favorite shows are ;)), it can be sad , it can be happy, it can be furious. There is so much more, I am pretty proud of the bot, and actually had offers from several companies that wanted to buy it (why I don't know). This was just a project to test the power of mIRC scripting but it entertained many people for many months. Right now I don't have him running, but if people are interested , I can start him up someday and show it.

Turing test

Yeah, this proposal is AI complete [tuxedo.org]. As a matter of fact, it is more or less exactly a Turing test.

But then, the Turing test can be passed with some success on unsuspecting examinators, for some time at least. When the Doctor program was written (see for example M-x doctor under Emacs), some people were fooled for some time. IRC would be a likely place to fool people. Usenet - or Slashdot - even more so, because the conversation delay makes it even easier to stick a lot of coined phrases without being spotted out. (How many karma points could a Slashdot robot collect? That's an interesting question.)

Douglas Hofstadter, in ``Metamagical Themas'' has an interesting example of a Turing test transcript, in which he was almost fooled - not quite the way you'd want it, but the transcript is really fun reading.

--
Assertion "signature!=NULL" in ai/output/slashdot.c at line 1729 failed (core dumped).

Group chat dynamics

It is doubtful that the bots will have an unending, constructive conversation. Either conversation will halt, or an infinite loop will occur (good bots, however, will try to prevent both of these circumstances).

HELLO!?! What kind of crappy design would that be? Any would-be-human IRC bot worth its salt would emulate the behavior of humans who are AFK for half an hour, then jump back in with a random non-sequitor like "damn, i love pizza" or "YES!!! MY PROGRAM COMPILED!" They would also have to have areas of "interest" where they will jump in the conversation even if it didn't involve them, or areas of "disinterest" where they will just stay completely out of the conversation until something interesting comes up again.

BTW, aren't people missing that crucial distinction? This is a group conversation -- it's not like you're expected to have an answer to each thing someone else types; in fact, you'd be considered very rude if you did.

This is in some ways more difficult, and in some ways simpler, than the Eliza-psychologist type bot. Any way you look at it, though, a convincing implementation would be very, very difficult.

Re:I can see it now...

And for those who are too lazy to follow the link, here's a transcript of my conversation with it...

Eliza: Hello. I am ELIZA. How can I help you?
Me: I am having problems with my children
Eliza: Who else in your family?
Me: No one else; just my children
Eliza: Who else in your family?
Me: Don't you listen?
Eliza: Oh, I listen.
Me: Well, so why did you repeat that question?
Eliza: Oh, I repeat that question.
Me: Yes, I know you did.
Eliza: You are sure?
Me: Yes, I'm positive. Are you suggesting I'm psychotic?
Eliza: You are sure?
Me: Do you think I imagined it?
Eliza: Oh, I think you imagined it.
Me: So now I'm psycho.
Eliza: Does it please you to believe I am psycho?
Me: I said I thought that you are saying that I am psycho.
Eliza: How long have you been psycho?
Me: I guess since I started talking to you.
Eliza: Do you wish to started talking to me?
Me: Now you're not making sense
Eliza: How long have you been not making sense?
Me: YOU! YOU! NOT ME, YOU IDIOT! YOU!
Eliza: Oh, I not you I idiot I.

At this point I was laughing too hard to continue...

The Keanu Reeves Bot

Obviously (look here [air0day.com] :) Keanu Reeves is easy to emulate with some FSM of this kind:

switch(state) {
case 1: return "Dude..";
case 2: return "Dude.";
case 3: return "Dude?";
case 4: return "Whoa.";
case 5: return "No way.";
case 6: return "Excellent!";
case 7: return "Uh.. rock on, dude.";
default: return "What?";
}